This release includes a major refactor and finally provides the ability to perform hubness reduced nearest neighbor search with torch tensors 🥳

Added

• Support for torch, when using Faiss
• More metrics for Faiss

Changed

• Simplified kneighbors API, i.e. no queries can be supplied anymore, since they need to come from the source anyway

Removed

• Autofaiss support was removed

Fixed

• Several efficiency problems when not using Hubness Reduction were addressed

Changed

• Loosen class-resolver dependency
• Simplify dependencies for extras

Fixed

• Avoid computing initial kcandidates when not using HubnessReduction

Fixed

• Loosen python dependency restriction

New in 0.4.2

• Upgrade class-resolver dependency

New in 0.4.1

• Fix joblib vulnerability by raising minimum required version

New in 0.4.0

• Added Faiss to enable fast hubness-reduced nearest neighbor search on the gpu

New in 0.3.3

• Patch IPython vulnerability by setting it to >=7.31.1

New in 0.3.2

• Fixed some version constraints of dependencies

New in 0.3.1

• Relaxed some version constraints of dependencies
• Enhanced documentation and some type hint problems

New in 0.3.0

• More possibilities to instantiate Kiez (thanks to @cthoyt )
• Single-source use simplified
• Enhanced documentation